Basics

• File: abstraction fo data blocks on disk
• File System: controls how data on disk are stored in retrieved

Hard Disk

• Seek Time: time required to move read/write head to wanted cylinder (~10ms)
• Rotational Delay: Average time to wait until required sector moves under the read/write heads (~4ms)
• Read / Write Time: tiem required to read/write a sector (~50 $\mu s$)

File Types

• Regular Files: stores user information
• Directories: system files, store structure of file system
• Character special files: model I/O
• Block special files: model disks

File Naming:

• each file has a file name for identification
• naming convention depend on the OS

File Structure

• byte-oriented:
• record-oriented:
• tree-structured:

Types of File Access

• Sequential Access: Bytes can be read out only in sequential order
• Random Access: out of order readout possible

File Operations

• create
• delete
• open
• close
• read
• write
• append: can only add data at the end of the file
• seek: reposition file pointer inside file
• get attributes: retrieves relevant file attributes (access rights, time)
• set attributes
• rename

Directories (Folders)

• Files are usually organised in directories
• Directories are themselves also files
• hierarchical directories: files organised in user- and theme-specific groups

Pathnames

• can be absolut or relative to working directory (= current directory)
  • process can use syscall to change working directory
  • working directory is process-specific

Special File Names

• .: refers to working directory
• ..: refers to working directory's parent directory

Directory Operations

• create
• delete
• opendir
• closedir
• readdir: reads next entry from an open directory file
• link: creates link from existing file to directory
• unlink: removes a directory entry | if this only references a file -> file is removed

Disk Organisation

• Disk organised into Master Boot Record (MBR) and one or more disk partitions

Master Boot Record

• located in sector 0
• used to boot up computer
• contains partition table at the end
  • in PT at any time, 1 partition is marked as active

Disk Partitions

• start with boot block (No OS needed for this)
• layout depend on file system
• typically starts with superblock
  • contains key parameters of fyle system

File Allocation

• management of how blocks of filesa re allocated in physical memory
• File allocation Table (FAT)
• I(ndex)-Nodes

Allocation Types:

Contiguous Allocation

• file stored as contiguous run of disk blocks
• (+) simple, efficient
• (-) external fragmentation | How to reuse free blocks?
• (-) File Max Size must be known beforehand

Linked-list Allocation

• Blocks comprising file are stored as linked list
• First word of each block holds pointer to next block
• (+) avoids external fragmentation
• (+) directory entry only needs to store disk address of first block
• (-) random access very slow
• (-) block size not powr of two

Data Structures to Implement Allocation Types

File Allocation Table

• Contains pointers to subsequent file blocks
• only pointer to first block stored
• (+) block size can be power of two
• (+) random very fast
• (-) large FAT must be stored in memory

I-Nodes

• contains file attributes and addresses of disk blocks
• (+) i-node only needs to be in memory when file is opened
• (+) requires less memory

Directory Implementation

• Directory is file containing file attributes
• link to i-nodes containing file information

Shared Files

• files can appear in multiple directories with links
• hard links: link to i-node for the target file
• symbolic links:
  • special file specifying filename of target file
  • can also reference file in different partition
  • incurs additional overhead in filename resolution

Log-Structured File Systems (LFS)

• (+) improves efficiency for small disk writes
• small writes are problematic due too seek and rotational delays
• LFS combines small writes in buffer, writes them in one segment at end fo log
• treat disk as big circular log
• (-) seperate i-node map required to licate file i-nodes on disk
  • logging approach scatters i-nods all over, instead of fixed location
  • cleaner thread works through log to free disk space
    • removes obsolete data
    • copies valied data to memory to be written to new segment at end of log

Journaling File System

• (+) more robustness againts crashes
• system crash can leafe file system in inconssisten state
• approach:
  1. record operations to be completed in log
  2. write log to disk
  3. execute operations
  4. delete log => in case of crash log is still present. -> Perform operations in log

Virtial File Systems

• allows several different file systems on the same system
• defines standardised interfaces for file operations

Design Considerations

• Block-Size:
  • too large: waste disk space
  • too smale: files span over many blocks
• management of free blocks:
  • linked list of free disk blocks
  • bitmap

Disk Quotas

• limit on how many files (i-nodes) and disk blocks each user can use
• when file is extended by new disk blocks, checked againts quota record
• soft limits can be temporarily exceeded
  • warning issued, too many issued -> logging denied
• hard limits can not be exceeded